• Open Access

A multi-criteria based review of models that predict environmental impacts of land use-change for perennial energy crops on water, carbon and nitrogen cycling


Correspondence: Amy R. C. Thomas, tel. + 44 01 60 359 1833 e-mail: amy.thomas@uea.ac.uk


Reduction in energy sector greenhouse gas GHG emissions is a key aim of European Commission plans to expand cultivation of bioenergy crops. Since agriculture makes up 10–12% of anthropogenic GHG emissions, impacts of land-use change must be considered, which requires detailed understanding of specific changes to agroecosystems. The greenhouse gas (GHG) balance of perennials may differ significantly from the previous ecosystem. Net change in GHG emissions with land-use change for bioenergy may exceed avoided fossil fuel emissions, meaning that actual GHG mitigation benefits are variable. Carbon (C) and nitrogen (N) cycling are complex interlinked systems, and a change in land management may affect both differently at different sites, depending on other variables. Change in evapotranspiration with land-use change may also have significant environmental or water resource impacts at some locations. This article derives a multi-criteria based decision analysis approach to objectively identify the most appropriate assessment method of the environmental impacts of land-use change for perennial energy crops. Based on a literature review and conceptual model in support of this approach, the potential impacts of land-use change for perennial energy crops on GHG emissions and evapotranspiration were identified, as well as likely controlling variables. These findings were used to structure the decision problem and to outline model requirements. A process-based model representing the complete agroecosystem was identified as the best predictive tool, where adequate data are available. Nineteen models were assessed according to suitability criteria, to identify current model capability, based on the conceptual model, and explicit representation of processes at appropriate resolution. FASSET, ECOSSE, ANIMO, DNDC, DayCent, Expert-N, Ecosys, WNMM and CERES-NOE were identified as appropriate models, with factors such as crop, location and data availability dictating the final decision for a given project. A database to inform such decisions is included.


The European Commission (EC) prescribes an increase in renewable energy generation, of which biomass currently contributes 66%, to combat rising energy demand, climate change and overreliance on non renewable energies (European Commission, 2008a,b). Biomass can contribute to base load, since timing of generation is controlled, giving advantages over intermittent renewables such as wind and solar (Dondini et al., 2009). de Wit & Faaij (2010) suggest that, by 2020, 10.9 EJ yr−1 could be provided by traditional annual arable crops, and 21.8 EJ yr−1 from perennials. This increase in perennial cultivation will necessitate associated land-use change. Popular perennials include trees suitable for short rotation coppicing, such as willow or poplar, and rhizomatous grasses, such as switch grass, reed canary grass and Miscanthus (Powlson et al., 2005).

Changing land use and management affects the greenhouse gas (GHG) balance between the soil and atmosphere. Depending on previous land use (removal of existing cover and disruption of previously untilled soil) and on bioenergy crop type and its management (tillage, irrigation and N fertilizer requirements), GHG debt payback periods can last from 0 to 1000 years (Fargione et al., 2008; Kim et al., 2009). The EC (European Commission, 2008b), lists reducing GHG emissions among the core goals of bioenergy, hence it is crucial to identify possible emissions from land-use change and compare these to the appropriate fossil fuel reference system (Sims et al., 2006; Firbank, 2012). The IPCC synthesis report (IPCC, 2007) states that emissions of long-lived GHGs, which it lists as CO2, N2O and CH4, and halocarbons are the most significant.

Since land-use change contributes significantly to CO2, and agriculture is one of the main causes of N2O and CH4 generation [contributing 58% and 47% of anthropogenic emissions, respectively, making up 10–12% of total anthropogenic GHG emissions (AllaLi et al., 2007a,b; Malça & Freire, 2012; Smith et al., 2007, 2008)] it is important to consider the impact of land-use change and variation in agriculture on all three gases. Perennial crops differ from most frequently grown arable crops in terms of management requirements and growth patterns. Any change to these affects water cycling and availability, as well as soil physical and chemical properties and resulting carbon (C) and nitrogen (N) cycling and emissions (Haughton et al., 2009; Rowe et al., 2009).

Multi-criteria decision analysis

This research sets out an objective approach to identify the most appropriate assessment method of the environmental impacts of land-use change for perennial energy crops, in terms of carbon, nitrogen and water cycling. The approach takes the form of multi-criteria based decision analysis (MCDA) which is a robust, logical approach to meeting multiple demands. The term MCDA covers a wide range of approaches to structured decision-making, using ranking and multiple stakeholders, or individual decision-makers as appropriate. MCDA is comprised of four stages: structuring and framing the decision problem; articulating preferences; aggregating alternative preferences; and making recommendations. The relative importance of criteria must be considered, and this may incorporate weighting and ranking, or a compensatory approach where poor performance on one criterion can be balanced by good performance on another criterion (Guitouni & Martel, 1998). Alternatively, where several criteria are essential, a conjunctive method, as used by Hwang & Youn (1981) and Chen & Hwang (1992), may be necessary, where all approaches not meeting these key criteria are ruled out. A wide range of other approaches to MCDA have been developed for different situations. A comprehensive review of MCDA is provided by Guitouni & Martel (1998).

The objectivity of the MCDA approach is dependent on the objectivity of the criteria identification in the first two stages, so it is crucial that a transparent approach is used, based on appropriate stakeholder input, or data and literature. MCDA often incorporates stakeholder input for criteria identification; however, a literature review may be used in proxy. Therefore, the first stage is to identify what potential impacts have been discussed in the literature, and use these to map out criteria to be represented by the chosen approach. The decision structure is then used to outline the relative benefits of different assessment approaches, and the relative importance of different factors to be included in the assessment according to the second stage of MCDA. These considerations are then aggregated to select an appropriate MCDA approach, and an appropriate set of criteria on which to base recommendations.

Potential impacts of land-use change to perennial energy crops

Crop growth involves C fixation and associated gas exchange with the atmosphere, water and N uptake from soil and organic matter inputs from a plant to soil. These crop processes combine with agricultural management practices to affect hydrology, and soil and atmospheric composition. Therefore, crop growth, soil nutrient cycling and soil-atmosphere exchanges can be viewed as components of the same system, which mustbe represented in its entirety to make relevant predictions (Smith et al., 1997). The interrelationships between soil properties and crop growth and management, which produce outputs of soil-atmosphere gas exchanges, along with evapotranspiration (ET) and yield are illustrated in 1. These key outputs are likely to be altered by changes to the system.

Atmospheric CO2 is used by crops in photosynthesis and, in the case of energy crops, the C stored in harvested parts will be released on their combustion. However, non-harvested C will move into soil storage in the form of litter, roots and exudates (Powlson et al., 2005). This storage is only temporary. CO2 is also emitted from SOM by soil respiration, which can also be accelerated by erosion (Keoleian & Volk, 2005; Fargione et al., 2008). Separation of soil organic matter (SOM) into storage pools of microbial, litter and labile and passive humus, is significant, since these differ in rates of decomposition and availability (Porporato et al., 2003; Parton et al., 2010). The balance of carbon storage varies with land use, e.g. oil seed rape releases 0.24–0.40 t C ha−1 yr−1 depending on management; whereas Miscanthus stores 0.62 t C ha−1 yr−1; willow short rotation coppice (SRC) stores 0.09–0.18 t C ha−1 yr−1; and forest stores 0.32 t C ha−1 yr−1 (Brandão et al., 2011). Due to coupling of Cand N cycles, this will also affect N storage and emissions.

Nitrous oxide, N2O has a global warming potential (GWP) of 310 relative to CO2, so a change in emissions is particularly significant (Solomon et al., 2007). Soil N2O emissions vary according to rates of nitrification and denitrification, including proportions of which gases are produced, and rates of diffusion out of soil (Del Grosso et al., 2000; Li et al., 2005b). Nitrification and denitrification can occur concurrently, in aerobic and anaerobic soil pores respectively (Boyer et al., 2006; Chen et al., 2008). Controls on nitrification and denitrification have been identified as: water filled pore space (WFPS); diffusion and temperature, which control microbial activity and diffusion of products from soil (Müller et al., 1997; Chatskikh et al., 2005) as well as oxygen, labile C and inorganic nitrogen available to microbial biomass (Müller et al., 1997; Davidson et al., 1998; Li et al., 2005a,b; Verchot et al., 2006; Parton et al., 2010); and pH (Li et al., 2005a,b). These variables interact, for example microbial decomposition of C and N consumes soil pore oxygen (Boyer et al., 2006). Microbial population dynamics may also be regarded as a control, although they correlate with other identified variables (Müller et al., 1997; Del Grosso et al., 2000). Besides variation between sites, these controls will vary significantly over a field, particularly with factors affecting water movement such as topography and soil texture and compaction. As a result, nitrification and denitrification processes and associated emissions are highly spatially variable (Li et al., 2005a,b; Boyer et al., 2006). Processes also have significant temporal variation with ambient temperature change, precipitation and soil disturbance, nutrient input or uptake events (Grant et al., 2006; Beheydt et al., 2007). Crop and land management factors, which might affect the N cycling controls identified above include: tillage; the amount and type offertilizer N inputs; and crop N uptakes and removal of crop residue and litter (Delgado et al., 2010). Local climate factors may also affect C and N cycling processes, for example, freezing temperatures reduce oxygen diffusion into soil, can kill microbes and reduce available soil water (Frolking et al., 1998).

Perennials tend to have high nutrient use efficiency, limited yield response to fertilizer and higher associated soil nutrient retention, when compared to traditional annual arable crops. This means there is little incentive to use large quantities of fertilizer, and often virtually no requirement for nutrient inputs after the establishment season (Beale & Long, 1997; Christian & Riche, 1998; Boehmel et al., 2008; Culman et al., 2010; Heaton et al., 2010). Lower agrochemical requirements give perennial energy crops potential to reduce N2O emissions compared to annual arable crops (Cherubini et al., 2009). However, although N fertilizer inputs are often considered to have strong positive correlation with N2O output, their importance may be overstated, and tillagecan be a more useful determinant (Li et al., 1998). Itis therefore more accurate to consider individual situations, and interaction with tillage, than apply default emission factors (EFs) to predict emissions based directly on fertilizer values (IPCC, 2006; Chamberlain et al., 2011).

Tillage increases erosion and aerates soil increasing decomposition rates, and results in CO2 loss from soil. Therefore, no till practice for perennials should encourage soil organic carbon (SOC) accumulation producing a CO2 sink, as well as increasing soil pore oxidation of CH4 (Ball et al., 1999; Keoleian & Volk, 2005; Li et al., 2005a,b). This practice will affect the C component of both the ‘SOM’ and ‘GHG balance’ outputs identified in Fig. 1. However, this CO2 sink may not offset emissions from disruption of the previous ecosystem, and may not (at equilibrium) exceed the previous land use. Hence, C storage in the previous ecosystem must be taken into account (Fargione et al., 2008). Other impacts of reduced tillage include increased soil water-holding capacity, reduced soil erosion and reduced fossil fuel use by farm machinery (Li et al., 2005a,b).

Figure 1.

Interaction between crop growth and management and soil chemical and physical properties.

Since C and N cycles are coupled, it has been suggested that an increase in soil C with no till practice may stimulate microbial activity and release of N2O (Ball et al., 1999; Li et al., 2005a). Significantly, Li et al. (2005a,b) found that the GWP of N2O emissions from reduction in tillage offset soil carbon storage by 75–310%. This response of C and N cycling to land-use change is complex. For example, further studies found a decrease in N2O emissions with no till practice (Del Grosso et al., 2008; Kavdir et al., 2008; Regina & Alakukku, 2010). Variation in response has been noted according to previous land use and associated C and N accumulation, fertilizer inputs (Novoa & Tejeda, 2006; Hellebrand et al., 2008), soil type (Rochette et al., 2008), humidity (Regina & Alakukku, 2010), soil moisture, climate, soil physical properties and topography (Li et al., 2005a,b). There is also interaction with new land management and crop factors, such as fertilizer application schedule and leaf litter inputs. Prediction of these impacts is hampered by an incomplete understanding of processes. It is, therefore, crucial to consider both carbon and nitrogen cycling to identify GHG emissions, and site-specific variation in nutrient cycling changes must be anticipated. The N component of the ‘SOM’ and ‘GHG balance’ outputs identified in Fig. 1 is, therefore, also significant.

Wagener & Gupta (2005) state that empirical or statistical models based on input-output data relationships can perform better when representing poorly understood processes. By simulating relationships between data, such models may make accurate predictions for similar situations, however, since processes are not represented directly, performance decreases when applied to new situations. Statistical or empirical models are therefore inappropriate for scenario analysis where differences between sites are significant, or where predictions are required for future climates or differing atmospheric composition (Frolking et al., 1998; Groffman et al., 2000; Schoumans et al., 2009). Process-based models show better performance for scenario analysis (e.g. in the EUROHARP studies; Perrin et al. (2001) and Schoumans et al. (2009), provided that processes, and process interactions, are represented appropriately. This explicit process representation increases input data demand for model execution (Keoleian & Volk, 2005; Clifton-Brown et al., 2007; Chen et al., 2008).

Given spatio-temporal variation in controlling processes, model representation can be expected to perform better at an appropriate resolution, incorporating data on this variation. Although topographical data to inform lateral water movement through soil pores is usually available, a lack of distributed data on soil texture and chemical composition may necessitate use of interpolated or average values (Potter et al., 1996; Chen et al., 2008; Schoumans et al., 2009). Small-scale variations in processes may be unnecessary for the annual field scale level of representation required for identifying the impacts of change in land use (Grant & Pattey, 2003; Del Grosso et al., 2005; Del Grosso et al., 2008). Without representation of spatial variation in controlling factors, it has been suggested that explicit representation of microbial processes will not improve model performance (Potter et al., 1996; Chen et al., 2008).

Scale and resolution must be considered in three dimensions with controls on C and N cycling also varying with depth, given that deeper soil layers may make an important contribution to N2O emissions (Rochette et al., 2008; Bessou et al., 2010). However, representing variation over the profile requires site-specific soil data for each layer, significantly increasing data demand. To identify whole agroecosystem emissions, it is crucial to consider both direct N gas emissions from soils, and indirect emissions from leached or stored nitrates. The IPCC emission factor (EF) for leached N is twice that for applied N, consequently an increase in soil emissions could be offset by a decrease in leaching and downstream indirect emissions (Groman et al., 2000; Nevison, 2000). These indirect emissions must be regarded a key component of the ‘GHG balance’ output identified in Fig. 1. Therefore, some representation of the soil profile is required for representation of leaching and associated indirect emissions (Chen et al., 2008).

Evapotranspiration (ET) also varies with land use, and there are significant differences between perennials and annuals. Key factors are growing season (which dictates the timing of water usage and interception) (Goodrich et al., 2000; McKendry, 2002; Vanloocke et al., 2010), rooting depth (deeper roots may enable water access during shortages) (Finch et al., 2004) and photosynthesis type [C4 plants have double the water use efficiency of plants using C3 photosynthesis (Smeets et al., 2009)]. Although photosynthesis type is important, a C4 perennial such as Miscanthus may cause greater evapotranspiration than a C3 annual (McKendry, 2002; Smeets et al., 2009), due to high productivity and long growing season increasing transpiration, and increased interception and evaporation (McKendry, 2002; Vanloocke et al., 2010). Vanloocke et al. (2010) suggest that rates of ET will also vary between sites. The significance of change in ET is also site specific, such that reduction in effective rainfall may be detrimental in some locations, or may reduce flood risk in others (Rowe et al., 2009). ET is identified as an altered system output in Fig. 1.

Yield is also identified as an output on Fig. 1; it is significant in terms of economic viability, as well as dictating the land area required for a given amount of energy. Therefore, it is important to consider potentially restricting factors such as water availability, climate, soil type and nutrient availability (Aylott et al., 2008; Richter et al., 2008; Lovett et al., 2009).

The environmental impacts of cultivation of the new crop as well as the removal of existing biomass and the interaction of a new crop and management regime with the soil conditions produced by previous land use must be considered (Fargione et al., 2008; Searchinger et al., 2008; Brandão et al., 2011; Nikièma et al., 2012). After change in land use, there is a variable period before a catchment reaches steady state due to: nitrogen retention in soils and aquifers (which can last for decades to centuries) (Haag & Kaupenjohann, 2001; Van Breemen et al., 2002); time taken for establishment of new vegetation (canopy, roots etc.); and rates of processes, such as soil organic matter accumulation, decay or erosion (Van Breemen et al., 2002; Keoleian & Volk, 2005; Darracq et al., 2008). SOM accumulation rate for a new land use will vary over time; it is often greatest during establishment and usually curvilinear until SOM reaches equilibrium, under constant conditions (Smith, 2004; IPCC, 2006). For useful prediction, the assessment approach must, therefore, take into account both management and properties of the new crop, and legacy of the previous land use, as well as how these interact to affect the whole agroecosystem, and what timescale needs to be represented. This may require a capacity to predict SOM status based on previous land use and management, where measured data are unavailable.

The decision problem is therefore structured as a need to predict site-specific changes in processes and resulting GHG balance and ET, for land-use change to perennials, taking into account the variables discussed above. It is useful to consider whether or not existing assessment methods for land-use change fulfil this need, and what approaches might be appropriate.

Existing assessment requirements before land-use change

In the EU, there is variation in the level of permits required for a bioenergy project, and no universal requirement for prediction of environmental impacts of associated land-use change (IPCC, 2006). However, there is a standard set of IPCC guidelines set out for calculating the GHG impacts of land-use change (IPCC, 2006). In the United Kingdom, environmental impact assessment (EIA) is required for crop cultivation on natural or semi natural land, or land that has been uncultivated for 15 years (Natural England, 2009). Woody perennial energy crops fall under the Forestry Commission 1999 regulations requiring environmental impact assessment for afforestation on designated sites, or for plots over 2 hectares on all other land (Natural England, 2009). Energy Crop Scheme (ECS) grants are available from Natural England to cover 50% start-up costs of perennial cultivation for energy; as part of this decision some environmental appraisal is undertaken (Natural England, 2009). From summer 2011, this included application of the IPCC methodology to predict GHG emissions from land-use change and the new agroecosystem.

The IPCC methodology for calculating GHG from land-use change is a three-tiered approach. Tier 1 requires least data, applying previous C storage assumptions based on ecosystem type, and biomass accumulation rates according to agroecosystem and climate region. Tier 2 improves on Tier 1 by application of country-specific rates for new biomass accumulation, and accounting for incomplete removal of previous biomass. Tier 3 is most data intensive, and applies measured data on existing biomass and calculates accumulation rates based on species and management-specificfactors, local climate and soil and process-based models (IPCC, 2006). Similarly the IPCC methodology for emissions from N inputs has three tiers, building from generalized emissions factors (EF) to more process-based modelling for Tier 3, for which data may be difficult to acquire. Site-specific variation in response to land use, and land-use change outlined in Section 'Potential impacts of land-use change to perennial energy crops' is not incorporated into Tiers 1 and 2 of the IPCC approach.

To minimize data requirements and costs, in most cases either Tier 1 or 2 is applied, with Tier 3 only applied where the land classification has been identified as significant for changes in C stocks (IPCC, 2006). However, current understanding of crop-soil processes is incomplete, making such change difficult to anticipate, and furthermore change in N may be more relevant in terms of GHGs. Due to the data requirements of process-based models advocated by the IPCC Tier 3, it may be impractical to perform modelling for all potential sites. However, the application of such a model at a range of sites, combined with appropriate assessment of the findings, may improve current understanding of site-specific factors controlling GHG impacts of land-use change.

The IPCC land-use change assessment focuses on GHG emissions, meaning that change in ET for a new crop is not considered, although it could be an output for a Tier 3 process-based model, and would be incorporated by a full EIA. Since there is no accepted EU-wide framework for prediction of the environmental impacts of land-use change for energy crops, then a change in ET may be considered in some cases, but not in others. Economically, it would be unwise for perennial energy crops to be cultivated where irrigation might be required, or where yields might be limited by water availability, and therefore some level of consideration by a farmer is likely before land-use conversion. This consideration may be restricted to economics, and may not consider ecosystem impacts where it is not obligatory.

Model selection for predicting impacts on water and GHG balance of land-use change to perennial energy crops

Since cultivation of perennials for energy is likely to increase, and may represent a change in an agroecosystem, it is useful to predict potential impacts before land-use change, particularly to assess whether or not the GHG mitigation objective identified in Section 'Introduction' is being achieved. Preferred approaches to the decision problem structured in Section 'Multi-criteria decision analysis' are discussed and justified here.

Models are a relatively cheap and convenient way of predicting and optimizing impacts of planned changes, and process-based models perform best for scenario analysis for agroecosystem change where there is no site-specific calibration data (Frolking et al., 1998; Groffman et al., 2000; Schoumans et al., 2009; Chirinda et al., 2010). Data demands may be significant, so for large scale or global assessments, empirical models are more suitable (see Hillier et al. (2012) for a recent assessment and application).

In order for a process-based model to achieve good performance, representation of processes and their interaction must be as comprehensive as possible. Therefore, the next step is to identify a process-based model which can represent the impacts of land-use change for perennial energy crops, as outlined in Section 'Potential impacts of land-use change to perennial energy crops', and can be applied at individual sites where land-use change is planned, perhaps as part of a Tier 3 IPCC approach. Required model outputs are as indicated in Fig. 1:

  • GHG balance (including indirect emissions from leached nitrates)
  • SOM status
  • Yield
  • Change in ET

A conceptual model is presented (Figure 2) to illustrate the complex water and nutrient cycling processes which need to be represented, for ease of matching to the chosen model.

Figure 2.

Agroecosystem conceptual model: C, N and water cycling at plot scale.

An important review by Chen et al. (2008) makes a general assessment of N2O emission model characteristics, intended to inform researchers on how to select a model suitable for their purposes. The current study builds on this, and similar reviews, by making specific assessment of model capability to represent a perennial agroecosystem, and predict impacts of land-use change and cultivation, in terms of the outputs listed above. Overall, there is no existing comparative assessment of model (or sub model combination) suitability to predict these outputs and assess impact of land-use change for energy crops.

The whole agroecosystem approach, advocated from assessment of Fig. 1, is not common in research on land-use impacts, which tends to focus on a single output (for example Grant & Pattey, 2003; Richter et al., 2008; Toma et al., 2010; Vanloocke et al., 2010). This may be because agricultural GHG emissions and evapotranspiration are separate issues in terms of impact scale; evapotranspiration is likely to impact local water resources and ecosystems (de Fraiture & Berndes, 2008; Vanloocke et al., 2010); whereas GHG emissions are a global issue in terms of climate change (Dondini et al., 2009). However, the whole agroecosystem approach is well recognized in agriculture, and is often utilized for improving efficiency of amounts and timings of fertilizer inputs and irrigation.

For harmony between studies, many model reviews advocate application of the same model at different sites (e.g. in the EUROHARP studies; Perrin et al., 2001; Schoumans et al., 2009). However, it is difficult to find a universal model which provides a similar level of accuracy at all sites; models tend to perform better in a similar environment to where they were developed, and may sometimes fail to include processes occurring at new sites such as snow melt and fluvial retention (Frolking et al., 1998; Groffman et al., 2000; Schoumans et al., 2009). Therefore, a geographical scope must be outlined for each model, in terms of development location, and process representation, to identify the availability of a potential universal model.

Besides affecting soil within an altered plot, land-use change will affect areas down slope and downstream of the plot (Lane et al., 2009). Land-use changes are most conveniently considered at plot scale due to spatial variability in current land use and its vintage. Therefore, to reduce data requirements and computational demands, it is logical to model at plot scale and make a separate consideration of impacts down slope and downstream.

As mentioned in Section 'Potential impacts of land-use change to perennial energy crops', SOM-building processes are temporally variable, hence the first year's model results will not be representative of a full cycle average, and it is useful to run the model until equilibrium over a full growing cycle. Accumulation from a second growing cycle can be expected to offset emissions from decomposition of old roots (Keoleian & Volk, 2005).

Land-use change and energy crop cultivation impacts falling outside the scope of the physiological crop growth process such as ecology and biodiversity, landscape structure, socioeconomic systems and emissions from fuel use, will not be represented by an agroecosystem-based model (Paine et al., 1996; Hanegraaf et al., 1998; Cherubini, 2010). The predictive models considered are not intended to give a complete assessment, but could contribute a useful input to a broader lifecycle analysis (LCA).

Criteria for model selection

Multiple criteria must be taken into account in choosing a model, and selection requires a transparent, well-structured decision-making process, for clarity. This research aims to take an objective approach to identifying which models may be best suited to the purposes outlined in Section 'Model selection for predicting impacts on water and GHG balance of land-use change to perennial energy crops'.: relevant processes (see Fig. 2) and how they would be affected by land-use change for perennials (see Fig. 1); and appropriate resolution (see Section 'Model selection for predicting impacts on water and GHG balance of land-use change to perennial energy crops'). A list of models was compiled, and an MCDA methodology was applied, based on relative importance of the criteria identified.

Traditionally, MCDA incorporates weighing of criteria based on stakeholder input at this stage; however, this step is not appropriate here. As identified in Section 'Model selection for predicting impacts on water and GHG balance of land-use change to perennial energy crops', it is necessary to use a process-based whole agroecosystem model to make accurate predictions for new sites, and in new or future climate conditions. As illustrated in Fig. 1, crop soil and site properties interact to control output and process-based models are unlikely to perform well unless the whole system and all relevant processes are represented, which removes the needfor stakeholder ranking of process representation requirements (Del Grosso et al., 2005; Frolking et al., 1998; Groffman et al., 2000; Schoumans et al., 2009; Smith et al., 1997). To make useful predictions, it is also necessary for all outputs identified in Section 'Model selection for predicting impacts on water and GHG balance of land-use change to perennial energy crops' tobe produced. Therefore, crucial primary criteria to compare process-based models were identified as:

  • Representation of complete agroecosystem
  • Potential to meet all output
  • Explicit representation of required soil, crop and land management processes (‘component’ boxes in Fig. 1)

A two stage MCDA framework was adopted here, and using the conjunctive method, models not meeting all primary criteria were excluded at the initial assessment stage. Lesser or study-specific criteria were grouped in a second table (Guitouni & Martel, 1998). Indirect emissions do not affect representation of other processes, and could be represented by a separate sub model. However, the high IPCC indirect GHG EF suggests that failing to represent indirect emissions would give poor representation of the system impacts, and so the potential to meet this output was included as a primary criterion (Groman et al., 2000; Nevison, 2000).

Some processes affecting soil nutrient cycling were identified as of lesser importance by reviewing the literature. For example, microbial population dynamics correlate with other parameters and may be represented implicitly (Müller et al., 1997; Del Grosso et al., 2000), and the importance of freeze thaw is dependent on study location and climate. Therefore, models unable to represent these processes were not ruled out, but this ability was considered as a limiting factor.

Since resolution appears to be of secondary importance, this was included in a second criteria matrix, along with locations and crops to which the model has been applied (Grant & Pattey, 2003; Del Grosso et al., 2005; Del Grosso et al., 2008). These secondary criteria are of varying importance depending on the location, crops and data availability of the study in which a land-use change model is required.

Due to the required range of outputs, it may be necessary to couple models. In particular, indirect GHG emissions tend not to be a standard output for agroecosystem models. Potential impacts of coupling must be considered, since accuracy can be reduced depending on the quality and suitability of data feeding into the coupled model (Smith et al., 1997). Coupling representation of soil physics and biology is currently problematic, and improvements could significantly enhance representation of generation and transport of GHGs in soil (Blagodatsky & Smith, 2012). Model complexity may increase scope, however, simple models may perform better when coupled, and structure may be more important than complexity in determining model accuracy (Smith et al., 1997; Perrin et al., 2001). Although structure and complexity may be considered, validation performance is likely to be most indicative of suitability. Therefore, if the model has already been tested for specific perennials, then relevant process representation can be ascertained. It is important to base confidence in the model on performance at validation as opposed to calibration (Perrin et al., 2001).

Compiling a database of suitable models

There are a great number of models available which could be used to predict part(s) of the required outputs, and so it is not possible to consider every existing model. Due to the high GWP of N2O outlined in Section 'Introduction', and the uncertainty of N2O emissions response to change in land use and management outlined in Section 'Potential impacts of land-use change to perennial energy crops', it is sensible to focus on models with good potential in this area. A database of several models widely utilized by existing studies (for which it is, therefore, possible to establish reliability, scope etc.) was compiled from the literature. A Scopus search was run to find models using ‘TITLE-ABS-KEY-AUTH (soil N2O emission model)’, which gave 403 results. Of these 403 results, several models were featured strongly; 190 discussed the IPCC methodology, 112 DNDC, 54 DayCent, 34 Expert-N, 15 Ecosys, 15 Hole In The Pipe (HIP), 14 ANN, 13 NLOSS, 8 WNMM, 9 NOE (6 AS CERES-NOE), 8 CASA, 6 STICS, 5 FASSET, 4 InfoCrop, 3 ANIMO, 3 FullCAM, 2 MCROPS and MGRASS, 1 ECOSSE and 1 INCA.

Discussion of comparative model suitability

It is outside of the scope of this article to give detailed descriptions of all models. Descriptions can be found in IPCC (2006) for the IPCC methodology; Davidson et al.(1998) for HIP; Ryan et al. (2004) for ANN; Brisson et al. (1998) for TNT-STICS-NEMIS; Aggarwal et al. (2006) for InfoCrop; Potter et al. (1996) for CASA; Renaud et al. (2006) for ANIMO; Lokupitiya & Paustian (2006) for FullCAM; Roelandt et al. (2005) for MCROPS and MGRASS; Whitehead et al. (1998) for INCA; for ECOSSE Smith et al. (2010); and Chen et al. (2008) for DNDC, DayCent, FASSET, Expert-N, Ecosys, NLOSS, WNMM and CERES-NOE.

Of the models identified from the Scopus search; ANN, tiers 1 and 2 of the IPCC methodology, FullCAM, MCROPS and MGRASS were ruled out as non-process based. However, where the data demands of a process-based model cannot be met, one of these models may be applied if it has been calibrated at a similar site.Table 1 details comparison of remaining models on primary criteria, applying a conjunctive approach, as detailed in Section 'Criteria for model selection'.

Table 1. Primary criteria to select a model capable of representing impacts of land-use change for perennial energy crops
 Full agroecosystemOutputs met/ Potential to produce all outputProcesses modelled explicitly
  1. GHG- greenhouse gas; SOM- soil organic matter; ET- evapotranspiration.


Direct GHG emissions,leaching, ET, yield, SOM.20, 22, 35

Downstream N transformation can be separate or coupled.

Separate root and shoot growth, tillage, water balance, leaching, soil aquiferexchange and riparian zone circulationsoil N, C and temperature.20, 22, 35 Ruled out: No separation of slow and fast soilpools.
INCARuled out: Partial26 Direct and indirect N GHG emissions, ET, SOM.26Requires additional crop modelNitrification and denitrification, seasonalaverageplant uptake. Surface and subsurface flow pathways, river flow and N concentration and transformation26
FullCAM (CAMFor, CAMAg, GENDEC, Roth-C)Ruled out: Partial32 Direct C GHG emissions, SOM.32, 24Requires additional sub models forN, yield and leaching.Soil C turnover, separated slow and fast C pools, residue decomposition, tillage, afforestation/deforestation24
HIPRuled out: Soil system only16 Direct GHG emissions, SOM.16Requires additional sub models forET, yield and leaching.Nitrification, denitrification, biological assimilation and lumped non-biological retention reactions16
CASARuled out: Soil system only18 Direct GHG emissions, SOM.18, 19Requires additional sub models for ET, yield and leachingPlant carbon fixation, nutrient allocation, litter fall, separated slow and fast coupled C and N pools, N mineralization and immobilization are represented implicitly according to C cycling.19 No ecological succession, compaction, tillage,irrigation18
CERES-NOEFull2 Direct GHG emissions, SOM,ET and yield2 Nitrification, denitrification, separated slowand fast coupled C and N pools, watermovement and nutrient leaching. No methanebalance or gas diffusion2. NOE2 representscompaction and tillage.36

Direct GHG emissions,leaching, yield, ET, SOM.2, 10

Downstream N transformationcan be separate or coupled.

Freeze thaw, multiple crop layers, leaf and rootvertical distribution, tillage, simultaneousdenitrification and nitrification lateral flowssimulated from topography, microbialdynamics, separated slow and fast coupled Cand N pools8, 11, 33
NLOSSFull2, 28, 29

Direct GHG emissions, ET, leaching,yield, SOM.2, 28, 29

Downstream Ntransformation canbe separate or coupled.

Microbial dynamics, separated slow and fast coupled C and N pools, simultaneous denitrification and nitrification, water/gas/heat translocation, not tillage or freeze thaw.1, 21,13 Ruled out: no representation of freeze thaw or variation in tillage

Direct GHG emissions, ET, Yield,Leaching, SOM.11, 2

Downstream N transformation canbe separate or coupled.

Freeze –thaw, tillage, separated slow and fast coupled C and N pools, simultaneous denitrification and nitrification, microbial dynamics22, 7

Direct GHG emissions, leaching, ET,yield, SOM23

Downstream N transformation canbe separate or coupled.

Microbial dynamics, separated slow and fast coupled C and N pools, simultaneous denitrification and nitrification. No soil tillage or cold temps.9 Ruled out: no representation of variation in tillage
ExpertNFull6, 9

Direct GHG emissions, ET, Yield,SOM, Leaching.6, 9

Downstream N transformation canbe separate or coupled.

Freeze thaw, snow, tillage, separated slow and fast coupled C and N pools, simultaneous denitrification and nitrification, microbial dynamics1, 20, 31
FASSET (agroecosystem submodel)Full2, 11

Direct GHG emissions, leaching, ET,SOM, yield.2, 11

Downstream N transformation canbe separate or coupled. Codeavailable for coupling2

Grazing/cutting, separate root and shoot growth, tillage levels, separated slow and fast coupled C and N pools, freeze -thaw, simultaneous denitrification and nitrification, microbial dynamics.24, 25, 32, 15 Ruled out: poor representation of variation in tillage

Leaching, yield, ET, SOM,extended for direct GHGemissions.13, 31,


Freeze –thaw, snow melt, tillage, retention in surface waters, simultaneous denitrification and nitrification, microbial dynamics, can have external crop model.2, 14, 16
DayCentFull27, 7

Direct GHG emissions, ET, Yield,Leaching, SOM.27, 7, 8

Downstream N transformation canbe separate or coupled.

Freeze –thaw, snow, tillage, root turnover rates (FORCENT), simultaneous denitrification and nitrification, separated slow and fast coupled C and N pools. No microbial dynamics.17, 18, 19, 19, 23

Direct gas emissions, ET, Yield,Leaching, SOM.4, 5

Potential to couple to SWAT fordownstream transformation, codeavailable for coupling.2, 34

Freeze –thaw, cold temp (-5 activation),snow coverage, weeding, tillage, water/gas/heat translocation, simultaneousdenitrification and nitrification, microbialdynamics, separated slow and fastcoupled C and N pools1, 5

Direct GHG emissions, ET, Yield,Leaching, SOM.37

Downstream N transformation canbe separate or coupled.

Tillage, separated slow and fast coupled Cand N pools, simultaneous denitrificationand nitrification37
References1. (Nevison, 2000) 2. (Chen et al., 2008) 3. (Brown et al., 2002) 4. (DNDC User Guide, 2009) 5. (Li et al.,1992) 6. (Frolking et al., 1998) 7. (Davis et al., 2010) 8. (Parton et al., 1998) 9. (Stenger et al., 1999) 10. (Grant,1995) 11. (Chatskikh et al., 2005) 12. (Silgram et al., 2009) 13. (Renaud et al., 2006) 14. (Wolf et al., 2005)15. (Ryan et al., 2004) 16. (Verchot et al., 1999) 17. (Groffman et al., 2000) 18. (Potter et al., 1996)19. (Davidson et al., 1998) 20. (Brisson et al., 1998) 21. (Beaujouan et al., 2002) 22. (Gascuel-Odoux et al.,2010) 23. (Aggarwal et al., 2006) 24. (Lokupitiya & Paustian, 2006) 25. (Roelandt et al., 2005) 26. (Whiteheadet al., 1998) 27. (Del Grosso et al., 2005) 28. (Riley & Matson, 2000) 29. (Christensen et al., 2006) 30. (Li et al.,2005a,b) 31. (Hendriks et al., 2008) 32. (Richards, 2001) 33. (IPCC, 2006) 34. (Gassman et al., 2007)35. (Brisson et al., 2003) 36. (Bessou et al., 2010) 37. (Smith et al., 2010)

The HIP, FullCAM, INCA and CASA models were ruled out as not representing linked crop-soil systems. Of the remaining models, all can provide the specific required outputs of yield, direct GHG emissions, leaching and ET. In terms of required process representation, STICS does not separate SOM pools according to availability and rates of decay, and so was ruled out. All of the remaining models can represent precipitation, flooding and irrigation, crop uptake and growth and fertilizer and residue application, as well as simultaneous denitrification and nitrification, using partitioning based on WFPS or aerated fraction of soil (Frolking et al., 1998; Li et al., 2000; Renaud et al., 2006; Chen et al., 2008; Chirinda et al., 2010). NLOSS and InfoCrop are ruled out because they cannot represent variation in tillage [which is crucial to modelling differences in management (Li et al., 2005a,b; Del Grosso et al., 2008; Rochette et al., 2008)]. NLOSS also cannot represent freeze thaw [which is crucial to representation of winter denitrification in temperate climates (Frolking et al., 1998)]. FASSET is unable to represent variation in mineralisation rates and GHG balance with tillage (Chatskikh et al., 2005), and function parameters for denitrification/nitrification controls are derived from site-specific data, which may impede performance at new sites (Chen et al., 2008).

Of the remaining models, some have limitations in representation of certain processes which must be taken into account. DayCent, WNMM and CERES-NOE do not represent SOM below 20 cm, so conclusions cannot be drawn for changes at greater depth, and DayCent uses implicit representation of microbial dynamics which may impede performance (Müller et al., 1997; Del Grosso et al., 2000; Chen et al., 2008; Del Grosso et al., 2008).

Secondary criteria of resolution and the locations and crops to which the model has been applied were then considered for ANIMO, DNDC, DayCent, Expert-N, Ecosys, WNMM, CERES-NOE and ECOSSE in. Table 2 All these models are suitable for representation of land-use change for energy crops, however, they differ in terms of scales and processes represented, and in terms of validation history. These should be considered as they apply to individual studies. Preference should be given to models tested and developed for appropriate locations and crops, and of a resolution appropriate to study data availability and output requirements. For example, CERES-NOE has not been validated for perennials, which may make it less applicable.

Given the history of model development, whereby a model is designed to solve a specific problem in the context of specific (usually localized) sites, the duplication identified in the literature search is, to some extent, inevitable. The lack of a universal model applied to all crops and regions reduces harmony between predictions; however, the effect of applying a model outside of its suitable range will diminish performance, negating perceived harmony. Table 2 can be used to identify the potential for a universal model for all land-use change to energy crops. However, besides needing to represent all relevant processes, true harmony would also require the model to perform equally between sites. It could be questioned to what extent harmony of representation can be expected between sites and crops which behave differently, i.e. where different processes are occurring in the landscape, or where crops have different management and growth cycles. If many of theprocesses occurring differ, different aspects of the model may apply at different sites, and these components may not have similar performance. Therefore, a universal model may only provide universal representation of commonalities.

Table 2. Secondary criteria to select an appropriate model from those capable of representing impacts of land-use change for perennial energy crops
 Regions appliedCrops parameterized forResolution
ANIMOEurope3, 1 Grassland, woodland, annual crops.1(option to use external crop model)


ANIMO N-L cat 1m vertical/horizontal0.01 km2 (1 ha) up to 100 km23


USA, Canada,

Australia, New


Europe, China

and India4, 5

Grassland, forest, annual crops,perennials4, 6, 16

Daily input data and decomposition.Hourly denitrification handling4, 6

Region mode; heterogeneous polygons –use coarsest resolution datasetSite mode; lumped6


USA, Canada,

Australia, New

Zealand and

Europe4, 7

Miscanthus, grassland, forest,annual crops4, 8 Daily.4 Spatially lumped at plot scale9

Germany, UK,

USA and

Canada4, 10

Annual crops4 Daily.11 Spatially lumped at plot scale12
EcosysUSA and Canada4 Grassland, forest, annual crops4 Hourly.13 Spatially referenced (coupledwith GIS) e.g. 50m by 50m28, 13

China, Australia,

Korea and


Grassland, annual crops4 Daily/hourly4, 14 Spatially referenced(coupled with GIS)15. 100-1000 m2 commonfor regional11
CERES-NOEFrance, Puerto Rico2 Annual crops2 Daily2
ECOSSEScotland, France,Germany17, 18 Grassland and forest17, 18 Daily. Spatially lumped, input parameters asavailable for regional scale18
References1. (Wolf et al., 2005) 2. (Gabrielle et al., 2006) 3. (Schoumans et al., 2003) 4. (Chen et al., 2008) 5. (Brown et al., 2002)6. (DNDC User Guide, 2009) 7. (Abdalla et al., 2010) 8. (Davis et al., 2010) 9. (Del Grosso et al., 2005) 10. (Frolking et al.,1998) 11. (Li et al., 2007a,b) 12. (Stenger et al., 1999) 13. (Metivier et al., 2009) 14. (Li et al., 2005a,b) 15. (Li & Chen,2010) 16. (Gopalakrishnan et al., 2012) 17. (Smith et al., 2010) 18.(Bell et al., 2011)

Validation performance under relevant crop, soil and climatic conditions may be used to aid the final choice of model where studies are available. Although validation is often based on time series GHG emissions, the spatio-temporal variability of emissions at field scale complicates the process. Point measurements may not be representative of longer term changes over the field as a whole, and validation based on change in soil C and N at several points to give proxy values for gaseous emissions may be more reliable (Bessou et al., 2010; Del Grosso et al., 2005; Li et al., 2005b).

Summary and conclusions

This review meets a need for identification of suitable models for predicting impacts of land-use change, particularly for perennial bioenergy crops, and is novel in that the whole agroecosystem approach is not commonly applied to research on land-use impacts. The method taken here uses a form of MCDA to apply criteria identified from the literature on the impacts of changes in land use and management. Although criteria were identified by an individual decision-maker, the use of literature to support application of the conjunctive approach to key criteria should ensure robust objectivity. This approach excludes models not capable of whole agroecosystem representation of all required processes and outputs, enabling more detailed consideration of all suitable models.

Land-use change for perennial energy crops often represents major change to tillage regime, evapotranspiration, crop rooting depth and seasonality of land cover, with knock-on effects on soil chemical and physical properties such as texture, WFPS and available C and N, and resulting GHG emissions of CH4, CO2 and N2O. GHG emissions from agriculture are significant, contributing 10–12% of the anthropogenic total. Also, water usage may differ from previous vegetation, affecting the catchment regime, and potentially limiting crop yield. Hence, understanding these variables may help to minimize ecosystem impacts and to fulfil the emissions savings potential of bioenergy crops. Prediction of GHG emissions using process-based models of nutrient cycling should include representation of soil water dynamics and crop growth, to account for N uptake and N and C inputs to the soil, such that crop yield and evapotranspiration will be included in model output. GHG emissions, evapotranspiration and yield are all significant impacts of agriculture. Therefore, it makes sense to model and assess these together, and to consider all three factors for a site. Indirect emissions may not be included in outputs, but can be easily calculated from predicted leaching rates.

The critical parameters required to model water usage and crop yield for perennial energy crops were considered by the MCDA presented here and indicated that DNDC, DayCent, Expert-N, ECOSSE, Ecosys, WNMM and ANIMO are suitable agroecosystem models for predicting potential impacts, and cover a range of regions and crop types. Given the availability of models meeting the review criteria, it is not necessary to build a new model for application to perennial energy crops, although it may be useful to adjust existing models to refine relevant processes and eliminate any unnecessary input data requirements. Data requirements of process-based models are always high, although there is some variation between hourly and daily resolutions, and distributed or lumped approaches. Thus, the selection of one of the identified models, based on the criteria and data availability of a given study, and using the information in the database in Table 2 is advocated.

In conclusion, the literature compiled here indicated that WNMM, DayCent and DNDC have been applied over the widest geographical scope. However, the models identified are only as good as their performance at validation, and this must be considered both in model selection, and also in conclusions and decisions based on model output. Good performance over a wide range of geographical regions suggests good representation of processes, although additional work on validation in other regions would be required before any of these models could be regarded as geographically universal. Model performance may remain better in locations similar to the region of model development which could negate perceived harmony. Hence, it would be necessary to perform comparative validation studies with all eight models for tropical, dry, cool and temperate locations for both arable and perennial crops to identify which, if any, might be fully regarded as a universal model.

Reliable application of identified models for scenario analysis may be restricted to conditions and crops for which a model has been successfully parameterized and validated.

Additional validation studies could increase the geographical scope of some models, or highlight regions where they should not be applied. In terms of climatic regions, validation gaps are as follows: ANIMO, dry, cold and tropical climates; ECOSSE and WNMM, cold and tropical climates; Expert-N, CERES-NOE, dry and cold climates; and Ecosys and DayCent, tropical climates. DNDC has been tested in dry, tropical cold and temperate regions, although further validation studies are required to cover all extremes of these climates, and it must also be remembered that models cannot be tested for future climates, which may affect longer term predictions. In terms of crops, DNDC and DayCent are well tested over a useful range of crop types (annual and perennial arable crops, grassland and forest) and should therefore perform well for land-use change to perennial energy crops, although calibration and validation may be necessary to add specific new crop types. Expert-N may prove unsuitable for perennials, as it has not yet been calibrated, although calibration of this model could prove extremely useful, since it incorporates more complete representation of tillage.

Identified models are only as good as their process representation. The representation of tillage is particularly significant and often incomplete, for example DayCent represents an increase in decomposition immediately after tillage, but does not represent changes in bulk density for tilled vs. no tilled systems. Similarly, WNMM only represents organic matter mixing, ANIMO applies user-defined, as opposed to calculated values for bulk density, and Ecosys, CERES-NOE, ECOSSE and DNDC do not represent change in soil structure with tillage (Li et al., 1992, 2007a,b; Groenendijk et al., 2005). Expert-N gives better representation of tillage, as impacts on bulk density are represented as well as mixing, however, this model would require calibration for perennial crops (Stenger et al., 1999). Lastly, a model which has been calibrated for both perennial and annual crops may still not perform well for the transition between the two, where representation of relevant processes is incomplete. Ultimately, the development of models identified by this review to give more complete representation of key processes should yield improved performance at validation.


Funding for this research was provided by the UK Energy Research Centre.